# can4hou6joeng4/boss-agent-cli [Health: Active]

**Category:** 🏢 Workplace & Productivity  
**Repository:** https://github.com/can4hou6joeng4/boss-agent-cli  
**GitHub Stars:** 2157  
**Views:** 2  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/can4hou6joeng4-boss-agent-cli

## Description
BOSS Zhipin recruitment workflow for AI agents. 49 MCP tools for job search, welfare filtering, recruiter messaging, pipeline tracking, and resume optimization.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "boss-agent-cli": {
    "command": "uvx",
    "args": ["--from","boss-agent-cli","boss-mcp"]
  }
}
```

## Documentation

## What can4hou6joeng4/boss-agent-cli does

can4hou6joeng4/boss-agent-cli provides an agent-accessible workflow for BOSS Zhipin, using the registered `zhipin` platform. It covers both sides of the recruitment process: candidates can find jobs, organize local shortlists, contact recruiters, apply, and manage conversations; recruiters can inspect their identity, search candidates, recommend people, send greetings, manage applications, exchange resumes, handle messages, and manage jobs.

Job discovery includes keyword and city searches, job details, cached result review, welfare matching, and recruiter-activity filters. Welfare terms are matched with AND logic, while local scoring can sort results. Candidate records can include local tags and notes for comparison and funnel tracking.

The package also includes resume and job-search assistance. Its AI commands can analyze job descriptions, polish or optimize resumes, assess fit, suggest keywords, draft cover letters, prepare interviews, and coach communications. Local model access supports Ollama and vLLM endpoints that expose an OpenAI-compatible interface.

## How it works

The same workflows can be driven from the terminal, MCP, JSON input, run IDs, or the Python API. A human can start the terminal wizard, while an agent can inspect the self-describing schema with `boss schema` and invoke commands or MCP tools. Standard output uses an envelope containing fields such as `ok`, `data`, `pagination`, `error`, and `hints`, which lets calling software distinguish successful results from recoverable failures.

Long-running collection uses checkpoints, bounded retries, time, request, and detail budgets, and explicit stop or resume operations. Crawl state is stored in SQLite, with incremental JSON, CSV, or XLSX outputs. The shared workflow model allows tasks to be started, resumed, or stopped through compatible command and agent entry points.

## Setup and configuration

Install the package with uv and install the Chromium runtime used for user-initiated login and local exports:

```bash
uv tool install boss-agent-cli
patchright install chromium
```

For MCP, configure a host to launch `boss-mcp` through uvx:

```json
{
  "mcpServers": {
    "boss-agent": {
      "command": "uvx",
      "args": ["--from", "boss-agent-cli[mcp]", "boss-mcp"]
    }
  }
}
```

Run `boss login` before authenticated work, then use `boss status` or `boss doctor` to check the session and local browser connectivity. Configuration is stored in `~/.boss-agent/config.json`; available settings include logging, request intervals, batch greeting intervals, CDP address, export location, platform, and role. Crawl commands require the additional crawl dependency.

## Tools and capabilities

- Job search, details, cached results, city lookup, history, welfare matching, and recruiter-activity filtering.
- Local shortlists, favorites, tags, notes, presets, comparisons, and statistics.
- Candidate greetings, applications, chat, pipeline updates, and digests.
- Recruiter identity, candidate search, recommendations, greetings, applications, resumes, messages, replies, and job management.
- Resumable crawl runs with budgets, checkpoints, results, resume, stop, and export operations.
- Resume and AI assistance, including local-model configuration and smoke tests.
- Schema and tool-definition export for agent integrations.

## Limitations and notes

Only BOSS Zhipin is currently registered, under the `zhipin` platform identifier. Recruiter-side automation uses `zhipin-recruiter`. Authentication, account-risk, network, and platform-risk conditions produce structured errors and recovery actions; a workflow stops when a platform risk state is detected.

Browser reuse through an existing CDP session requires a locally trusted browser that has already been logged in. The CDP endpoint should not be exposed publicly. The current release no longer includes the former Browser Bridge daemon, browser extension, or bridge installation extra.

Crawl exports intentionally omit security IDs, job IDs, and recruiter fields by default. Privacy cleanup can remove crawl state, budgets, and exports. Platform capabilities that are not implemented return `NOT_SUPPORTED` rather than being treated as completed actions.

_Full upstream README: https://allmcps.com/mcp/can4hou6joeng4-boss-agent-cli/readme_